Abstract
This study focuses on the predictive maintenance of rotating machinery – a fundamental asset in industries such as manufacturing, energy production, and transportation. The problem addressed is the frequent occurrence of undetected faults, such as bearing defects and shaft bending, which can lead to unexpected downtime and significant maintenance costs due to the limitations of traditional diagnostic methods in complex, noisy environments. To overcome these challenges, an integrated framework was developed that combines advanced vibration analysis techniques (including wavelet transforms and matching pursuit) with a suite of state-of-the-art machine learning models, including Random Forest, Support Vector Machine (SVM), Gradient Boosting, Convolutional Neural Network (CNN), and Long Short-Term Memory (LSTM). This innovative approach, characterized by robust feature extraction and data-driven modeling capabilities, achieves fault detection accuracies of up to 97 %, distinguishing it from conventional solutions. The findings demonstrate that the improved accuracy and reliability of the proposed framework effectively address long-standing issues related to incomplete fault detection and downtime in maintenance processes. By providing a scalable, noise-robust solution, the study contributes to industrial systems through significant reductions in operational overhead and downtime, thereby maintaining core business operations at peak performance